fix: Fix the bug that null data can not filtered by null expr (#41135)

issue: #41063 
pr: #41124

Signed-off-by: sunby <sunbingyi1992@gmail.com>

internal/core/src/exec/expression/EvalCtx.h

@@ -85,6 +85,16 @@ class EvalCtx {
         bitmap_input_.clear();
     }
 
+    void
+    set_apply_valid_data_after_flip(bool apply_valid_data_after_flip) {
+        apply_valid_data_after_flip_ = apply_valid_data_after_flip;
+    }
+
+    bool
+    get_apply_valid_data_after_flip() const {
+        return apply_valid_data_after_flip_;
+    }
+
  private:
     ExecContext* exec_ctx_ = nullptr;
     ExprSet* expr_set_ = nullptr;
@@ -94,6 +104,9 @@ class EvalCtx {
 
     // used for expr pre filter, that avoid unnecessary execution on filtered data
     TargetBitmap bitmap_input_;
+
+    // for some expr(eg. exists), we do not need to apply valid data after flip
+    bool apply_valid_data_after_flip_ = true;
 };
 
 }  // namespace exec

internal/core/src/exec/expression/ExistsExpr.cpp

@@ -24,6 +24,7 @@ namespace exec {
 
 void
 PhyExistsFilterExpr::Eval(EvalCtx& context, VectorPtr& result) {
+    context.set_apply_valid_data_after_flip(false);
     auto input = context.get_offset_input();
     SetHasOffsetInput((input != nullptr));
     switch (expr_->column_.data_type_) {

internal/core/src/exec/expression/LogicalUnaryExpr.cpp

@@ -30,9 +30,11 @@ PhyLogicalUnaryExpr::Eval(EvalCtx& context, VectorPtr& result) {
         auto flat_vec = GetColumnVector(result);
         TargetBitmapView data(flat_vec->GetRawData(), flat_vec->size());
         data.flip();
-        TargetBitmapView valid_data(flat_vec->GetValidRawData(),
-                                    flat_vec->size());
-        data &= valid_data;
+        if (context.get_apply_valid_data_after_flip()) {
+            TargetBitmapView valid_data(flat_vec->GetValidRawData(),
+                                        flat_vec->size());
+            data &= valid_data;
+        }
     }
 }
 

internal/core/unittest/test_expr.cpp

@@ -9,6 +9,7 @@
 // is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express
 // or implied. See the License for the specific language governing permissions and limitations under the License
 
+#include <algorithm>
 #include <boost/format.hpp>
 #include <fstream>
 #include <gtest/gtest.h>
@@ -16744,6 +16745,8 @@ TEST_P(JsonIndexExistsTest, TestExistsExpr) {
         test_cases = {
             {{"a"}, true, 0b1111111000000100},
             {{"a", "b"}, true, 0b0000100000000000},
+            {{"a"}, false, 0b0000000111111011},
+            {{"a", "b"}, false, 0b1111011111111111},
         };
 
     auto json_index_path = GetParam();
@@ -16752,7 +16755,7 @@ TEST_P(JsonIndexExistsTest, TestExistsExpr) {
     auto vec_fid = schema->AddDebugField(
         "fakevec", DataType::VECTOR_FLOAT, 16, knowhere::metric::L2);
     auto i64_fid = schema->AddDebugField("age64", DataType::INT64);
-    auto json_fid = schema->AddDebugField("json", DataType::JSON);
+    auto json_fid = schema->AddDebugField("json", DataType::JSON, true);
     schema->set_primary_field_id(i64_fid);
 
     auto seg = CreateSealedSegment(schema);
@@ -16774,12 +16777,15 @@ TEST_P(JsonIndexExistsTest, TestExistsExpr) {
         static_cast<json_index_type*>(inv_index.release()));
 
     auto json_field =
-        std::make_shared<FieldData<milvus::Json>>(DataType::JSON, false);
+        std::make_shared<FieldData<milvus::Json>>(DataType::JSON, true);
     std::vector<milvus::Json> jsons;
     for (auto& json_str : json_strs) {
         jsons.push_back(milvus::Json(simdjson::padded_string(json_str)));
     }
     json_field->add_json_data(jsons);
+    auto json_valid_data = json_field->ValidData();
+    json_valid_data[0] = 0xFF;
+    json_valid_data[1] = 0xFE;
 
     json_index->BuildWithFieldData({json_field});
     json_index->finish();